there are no (useful) warnings if domain names collide, and an already running domain may also break in case of collisions

[calestyo]

Hey.

My dCache runs at 11.2.3 (including an instance of frontend). To test Dmitry’s b6e2a9f I’ve installed 12.0.1 on a single node, only starting a second instance of frontend.

I was able to curl from that but it's swagger already didn't work. A little while later I've noticed that the "main" frontend instance, which still runs on 11.2.3 went completely crazy:
frontend.log

Can it be that running two instances of it isn't supported? I couldn't find that documented in the book.

Cheers,
Chris.

[DmitryLitvintsev]

We run multuple frontend instances in production (4 in fact) w/o issues. I will have a look at your log.

[DmitryLitvintsev]

It looks like the core problem - failure to connect to 11111 . The rest is just noise.

Also by chance you are not setting

frontend.cell.name=

to the same string across many instances (altho the log seems to indicate just the failure to connect to 11111)

[calestyo]

It looks like the core problem - failure to connect to 11111 .

Which is however strange, cause our whole dCache cluster has it's own subnet, and netfilter simply allows everything on from any IP of that subnet.
Plus, I can actually connect from the said node:

root@lcg-lrz-dc44:~# telnet lcg-lrz-dcache0.grid.lrz.de 11111
Trying 2001:4ca0:0:201:a::...
Connected to lcg-lrz-dcache0.grid.lrz.de.
Escape character is '^]'.
��sr dmg.cells.nucleus.CellDomainInfo�g$5�L
                                           _domainNametLjava/lang/String;L_rolet"Ldmg/cells/nucleus/CellDomainRole;_versionq~L_zoneq~xptcore0~r dmg.cells.nucleus.CellDomainRolexrjava.lang.EnumxptCOREt11.2.3p

Connection closed by foreign host.
root@lcg-lrz-dc44:~# 

Also by chance you are not setting

I don't, but shouldn't the default frontend.cell.name=Frontend-${host.name} make sure it's unique?

[calestyo]

I found the problem… in part it’s me being an imbecile, but one could argue that dCache should safeguard against it (and imbeciles 😜️):

While the cells had different names due to the default of frontend.cell.name=, the domains had not (I simply copy&pasted the frontend config from my other host, forgetting to change the domain name).

So there were now to domains named frontend, but this basically broke both (the already running and the new one).

IMO, there should be a warning/error, and dCache should simply refuse to "register" domains with names that are already used, thereby safeguarding those that already run.

[calestyo]

btw: There's a similar issue with pool names...
The documentation says:

   # Must be unique within this dCache instance
   pool.name =

I tried putting pools with the SAME name in two DIFFERENT Java domains and they actually start up (and I can go in their cell in the admin interface).
Does that cause breakage? I mean what's about t_locationinfo, which would then use the same name for both?

It fails however, as expected, when I try to create two pools with the SAME name within the SAME domain, but I guess that's just the generic check that no duplicate cell names must exist in the same domain.